Optimizing Geographical Weights in GWR Model by Incorporating Neighborhood Characteristics

Abstract

Geostatistics, owing to its proficiency in modeling spatially correlated phenomena, has extensive applications across various areas within the field of geography. These applications include spatial analysis, climate and weather studies, environmental investigations, and urban planning and development, among others. The Geographically Weighted Regression (GWR) model is one of the widely adopted models in geostatistical analyses. The essence of GWR is to take into account the spatial proximity when considering the impact of neighboring points in regression analysis. Efforts have been made to refine the model's accuracy by improving how the influence of neighboring points is weighted, employing methods beyond simple Euclidean distance—like network analysis—to better capture the complexity of spatial relationships. Despite these studies, the optimization of these weights has not been extensively explored in the literature. This study proposes an advancement in the optimization of GWR by incorporating not just the physical distance but also the characteristics of neighboring points into the weight calculations. By integrating a novel factor—neighborhood characteristics—into the geographical weights, the model aims to enhance its precision in quantifying the impact of nearby locations. This can lead to a more accurate understanding of spatial variance and improve the explanatory power of GWR models.